From Tissue Archives to Liquid Biopsy: Transfer Learning for MicroRNA-Based Lung Cancer Diagnosis

Area of Science:

• Biomolecular analysis
• Cancer diagnostics
• Bioinformatics

Background:

• Serum microRNA (miRNA) liquid biopsy shows promise for noninvasive lung cancer diagnosis.
• Challenges include vast miRNA combinations and limited clinical serum samples.
• Existing methods struggle with domain adaptation from tissue to serum data.

Purpose of the Study:

• To develop an efficient serum-based miRNA classifier for lung cancer detection.
• To enable effective domain adaptation and knowledge transfer from tissue to serum data.
• To validate a transfer learning strategy for liquid biopsy.

Main Methods:

• A transfer learning strategy with feature space alignment was developed.
• Genetic algorithm-driven feature selection identified a 4-miRNA panel.
• Reverse transcription quantitative PCR (RT-qPCR) quantified miRNA expression in serum samples.

Main Results:

• A 4-miRNA panel (miR-139-5p, miR-10a-5p, miR-148a-3p, miR-30d-5p) achieved AUC > 0.98 in tissue classification.
• The transfer-learned model reached 91.5% accuracy and 92.2% sensitivity in clinical serum samples.
• The approach demonstrated effective domain adaptation and knowledge transfer.

Conclusions:

• The developed transfer learning strategy offers a cost-effective solution for high-accuracy liquid biopsy.
• This approach overcomes limitations of small, clinically annotated serum sample sets.
• It significantly advances noninvasive lung cancer diagnosis using serum miRNAs.